Key aspects of this assessment and certification at a glance:
| No. | Consideration | Answer at-a-glance |
|---|---|---|
| 1 | What problem does this certification help solve? | Enables organizations to demonstrate that they sufficiently mitigate cybersecurity threats to the AI technologies they have deployed. The focus of this certification is security for AI systems. |
| 2 | How is this a problem that HITRUST can help solve? | HITRUST has all components needed to enable IT cybersecurity assurances correctly, consistently, and at scale. It’s what we’ve been building for the last 17 years. These are the same building blocks needed to enable AI assurances to the same degree of quality and reliability. |
| 3 | Who can achieve this certification? | Providers of AI systems, including AI Application Providers and AI Platform Providers. This certification is not for organizations simply using AI systems deployed by others (in the same way a SaaS user organization can’t get a HITRUST r2 certification on behalf of its service provider). |
| 4 | What is this certification… not? | While cybersecurity of the AI system is absolutely a key risk that must be understood and addressed, it is not the only risk introduced when AI is deployed. Organizations who achieve this certification will still need to navigate the additional risk areas in the Responsible AI landscape (such as AI privacy, ethics, and transparency). |
| 5 | How does this relate to ISO/IEC 42001:2023? | This is a complement to, not a replacement for, ISO/IEC 42001:2023. Organizations need both an AI management system and cybersecurity controls and assurances related to AI. To help organizations who use both ISO/IEC 42001:2023 and this AI security assessment and certification, we’ve prepared a crosswalk / mapping. |
| 6 | What types of AI models qualify for certification? | Generative AI, predictive AI (i.e., non-generative machine learning), and even the older rule-based AI (i.e., expert systems). |
| 7 | Which AI system layers are considered? | It focuses on the added IT components unique to AI (e.g., the model, the AI platform, and any specialized AI compute infrastructure in use) in addition the overall IT platform components normally scoped into a HITRUST assessment. |
| 8 | What are the AI security requirements needed for certification? | Up to 44 AI security-specific HITRUST CSF requirements, depending on how the assessment is tailored. |
| 9 | Which AI security threats are considered? | 13 threats. Some of the threats in this document’s AI security threat register are novel (e.g., prompt injection), and others are well-known security threats that are exacerbated by the deployment of AI. |
| 10 | How long is the certification valid? | Matches that of the underlying HITRUST CSF assessment. Meaning, it is valid for 1 year if attached to an e1 or i1 assessment and for 2 years if attached to an r2 assessment. |
Post your comment on this topic.